# longwell-block-buzz-workspace-agents-nostr-2026-07-21

## Veille

**Block** announcement from **July 21, 2026**, signed by **Tyler Longwell**: **Buzz**, an *open source* and **self-hostable** channel-driven workspace where humans and agents share the same room — chat, search, automation, and **Git hosting** on a single server, built on **Nostr**, an open protocol for signed messages and portable identities. Opening thesis: *« Models can do the work now. Teams still need somewhere to do it together. The bottleneck moved from intelligence to coordination. »* Three engineering pieces. **(A) Agent identity.** The starting point is a refusal — to stop lending one's credentials to a bot: *« We have been letting bots play dress-up as us. It's weird. It's dangerous. »* Each agent gets **its own key**, its owner signs a **narrowly scoped authorization**, and the agent then signs its work with its own identity. The delegation cryptography is conventional; the design decision is less so: *« authorization does not erase authorship »* — the agent remains the author, its *credential* proving who authorized it and under what conditions. Immediate consequences: a leaked agent key is revoked without touching the human identity, and withdrawing the owner prevents the agent from reconnecting, with its active sessions needing to be terminated separately. **(B) Git on object storage.** The observation: *« In the past, Git has always had a convenient rate limiter: humans »* — a group of agents produces months of person-commits and CI in a single afternoon, with many simultaneous writers, on forges sized for human fingers. Buzz stores repositories as **immutable, content-addressed packfiles** plus a **single mutable manifest pointer**; a *push* writes the objects first, then advances the pointer via **conditional compare-and-swap**, that swap being the commit point — workspace events announce the change, they do not define it. The protocol is **specified in TLA+ and model-checked** (durability, reconstruction, concurrent pushes), with the bounded result depending on three explicit object-store guarantees, hence a **conformance suite** every backend must pass. **(C) Interoperability and privacy.** Claude Code, Codex, goose *« and any agent speaking Agent Client Protocol »* work inside Buzz; switching model or harness leaves the project's identity, permissions, and history intact. Telemetry and cancellation travel as ephemeral encrypted messages, memory and cost accounting as durable encrypted messages — *« the server sees routing metadata, not those payloads »*. Memory argument: *« A conventional forge preserves the diff and a green check. Buzz also preserves why the obvious fix was wrong. »* Anti-lock-in argument: if Buzz disappears, the identity and signed history remain verifiable, Git stays Git.

## Titre Article

Buzz!

## Date

2026-07-21

## URL

https://engineering.block.xyz/blog/buzz

## Keywords

Buzz, Block, agentic workspace, channel, channel-driven, self-hostable, self-hostable, open source, Nostr, open protocol, signed messages, portable identity, keypair, keypair, agent identity, per-agent key, delegated authorization, delegation cryptography, authorization does not erase authorship, authorship, authorship, key revocation, shared credentials, identity impersonation, bot dress-up, authorization scope, scope, active sessions, Agent Client Protocol, ACP, Claude Code, Codex, goose, harness, harness portability, multi-agent orchestration, agent frontier, swarm, cheap agents, coordination, bottleneck, human middleware, shared context, channel memory, ephemeral channel, signed merge decision, visible history, failure paths, Git forge, Git hosting, Git on object storage, immutable packfile, content addressing, manifest pointer, compare-and-swap, commit point, TLA+, model checking, formal verification, conformance suite, object-store guarantees, device pairing, QR, six-digit code, encrypted ephemeral messages, routing metadata, GPU sharing, peer-to-peer inference, anti-lock-in, history sovereignty, Tyler Longwell, engineering.block.xyz, github.com/block/buzz

## Authors

**Tyler Longwell** — *« Building multi-player AI at Block »*, auteur unique et signataire à la première personne. Publié le **21 juillet 2026** sur le blog Block Engineering.

L'auteur a construit le prédécesseur du produit qu'il présente : *« I helped build Block's first Slack-integrated agent… I added enough glue to incorporate as a small adhesives business »*. L'argumentaire part d'une liste de questions restées sans réponse dans ce montage — *Does everyone get a bot? If people share one, whose credentials does it use?* (le sondage interne où *« everyone volunteered someone else »*), que faire quand une équipe veut changer de modèle ou de runtime. Note de clôture : *« We wrote this post in a Buzz channel with our team and our agents. We wrote Buzz there too. »*

## Ton

**Profile**: product-launch engineering post, **practitioner** register, written in the first-person singular — rare for a corporate announcement. Audience: platform engineers and teams already orchestrating multiple agents and paying the coordination cost.

**Style**: section titles built on **bee puns** (*Hive to Survive*, *Headfirst into the Swarm*, *Bee Yourself*, *Git for the Hive Mind*, *Honey, I Saved the Context*) that mask a thoroughly serious progression: lived problem → substrate (Nostr) → identity → storage → memory → governance. Three traits:

1. **Humor as an accelerant, not as filler.** *« a sufficiently ambitious toaster »*, *« alt-tabbing between them like a day trader »*, *« Connecting the two casually is how your cool new feature becomes an archaeological project »*. Every joke carries a technical argument.
2. **The moral framing of the identity problem.** *« The usual way to authorize an agent is to give it your credentials and hope it doesn't embarrass you. »* / *« If I see your face, I want to know it's you talking to me. »* The text treats agent identity not as a compliance constraint but as a **matter of social trust**.
3. **The admission of immaturity, unhedged.** *« We're still very early in Buzz's life. There are rough edges and giant chasms between what we can see on the horizon and what exists today »* — placed **after** the argument, not buried inside it.

**Marker phrases**: *« The bottleneck moved from intelligence to coordination »*, *« authorization does not erase authorship »*, *« We have been letting bots play dress-up as us »*, *« Nobody enjoyed being middleware »*, *« We put Git on object storage. On purpose! »*, *« Buzz also preserves why the obvious fix was wrong »*, *« it's 2026: software got cheap. Taste didn't »*, *« A protocol anyone can rebuild is a protocol nobody can lock you into »*, *« The room itself has become part of the team's intelligence »*.

**Epistemic stance**: **assertive on design, silent on evidence**. No measurement, no benchmark, no adoption figures — the benefit claims (*« easier, faster, and more effective than any coordination tool or process we've used »*) are **internal-usage testimonials**, not results. The post's only *formal* proof covers **Git storage** (TLA+), not the claimed productivity. This is the exact complement to the series' next post ([[patel-block-buzz-teams-tokens-benchmarks-2026-08-06]]), which supplies the figures — and tempers the promise.

## Pense-betes

- **The sentence that justifies the category's existence**: ***« Models can do the work now. Teams still need somewhere to do it together. The bottleneck moved from intelligence to coordination. »*** Worth reusing verbatim to argue for an "agent platform" budget: the useful marginal spend is no longer in the model, it's in the **room where the work happens**. The associated diagnostic symptom is even more operational: *« Agents made individual work faster and teamwork slower »* — if your teams observe exactly that, the problem isn't your model choice.
- **The design choice worth stealing, even without adopting Buzz**: ***« authorization does not erase authorship »***. Each agent has **its own key**; the owner signs a **narrow authorization**; the agent signs its work **in its own name**, and its *credential* states **who authorized it and under what conditions**. Translated into three properties to demand from any enterprise agent platform: 1. **Isolated revocation** — a leaked agent key is revoked **without reissuing the human identity**; 2. **Unambiguous attribution** — the audit trail distinguishes "the agent did X" from "the person did X"; 3. **Cascading withdrawal** — a withdrawn owner leaves their agents unable to reconnect (+ active-session termination if the risk is immediate). **The named anti-pattern is very widespread**: sharing a human's credentials with a bot. Check what *your* current Slack/CI integration does before judging others'.
- **Git has lost its rate limiter — the observation transfers to your entire tooling chain.** *« In the past, Git has always had a convenient rate limiter: humans. We sleep, eat, attend meetings, and sometimes think before pushing. Agents remove those limits. »* A group of agents produces **months of person-commits and CI in a single afternoon**. **To do at home**: list the systems whose sizing implicitly assumed a human cadence — forge, CI, review, API quotas, alerting, security review — and see which breaks first. It's a capacity exercise, not a governance one.
- **The technical answer, clean and copyable**: repositories as **immutable, content-addressed packfiles** + **a mutable manifest pointer**; a *push* writes the objects first, then advances the pointer via **conditional compare-and-swap**. **The rule that keeps the model sound**: *« That pointer update is the commit point. Workspace events announce the change; they do not define it. »* — the channel event is a **notification**, never the source of truth. Pattern generalizable to any event-driven integration over object storage.
- **Read the formal guarantee with its scope clause.** The protocol is **specified in TLA+ and model-checked** (durability, reconstruction, concurrent pushes) — but *« the bounded result depends on three explicit object-store guarantees, so every backend must pass a conformance suite »*. In other words: **the proof holds for a compliant object store**, and compliance is delegated to a test suite. Question to ask before any self-hosting: *does my S3-compatible store pass the suite?* The post doesn't name the three guarantees.
- **Interoperability — the point that matters for tooling strategy**: **Claude Code, Codex, goose, and any agent speaking [[Agent-Client-Protocol]]** work inside Buzz; *« Change the model or harness and the project keeps its identity, permissions, and history. »* → identity and history are attached to the **project**, not the *harness*. Mind the acronym: this is indeed the **Agent Client Protocol** (client ↔ agent), not the *Agentic Commerce Protocol* — cf. [[girard-acp-deux-protocoles-un-sigle-2026-08-02]]. For the ACP building block itself: [[agentclientprotocol-introduction-2026-08-02]].
- **The real promised value isn't chat, it's the memory of reasons.** *« Search auth refresh six months later and find the report, rejected fix, patch, review, and final call. A conventional forge preserves the diff and a green check. Buzz also preserves why the obvious fix was wrong. »* → **the connection to make**: this is a *context engineering* argument disguised as a collaboration argument — the shared context stays in the channel **instead of being compressed into private prompts**. To connect with [[kb-context-engineering]] and the agent-memory thesis.
- **Observation worth verifying elsewhere**: *« The agents also invent coordination we didn't script: recruiting each other, splitting work into side channels, handing tasks across contexts. »* **Anecdotal and unmeasured** — but if it holds up, it's a major design fact (coordination emerges from an open messaging substrate, with no orchestrator). To be checked against the follow-up post's figures, which show the opposite instead — that **team composition doesn't pay off on short tasks**: patel-block-buzz-teams-tokens-benchmarks-2026-08-06.
- **Privacy — what's encrypted and what isn't**: live telemetry and cancellation = **ephemeral encrypted messages**; memory and cost records = **encrypted but durable**; the server sees *« routing metadata, not those payloads »*. Buzz can route inference to **another member's machine** (GPU pooling): Buzz introduces the authorized peers, then **encrypted model traffic flows directly between them**. → Interesting for a team wanting to pool local compute; worth reviewing on the DLP side, since the workload leaves the server perimeter.
- **The anti-lock-in argument, phrased for reuse**: *« If Buzz disappears, your identity and signed history still verify. Git remains Git and can be rehosted… A protocol anyone can rebuild is a protocol nobody can lock you into. »* And the governance version: *« Running the infrastructure shouldn't mean owning the identities, history, or work that passes through it. »* → **Transferable purchasing criterion**: for any agent platform, ask what **survives the vendor's disappearance** — identities, verifiable history, rehostable artifacts. A vendor's argument, but **testable**: Block publishes code, protocol specs, test vectors, security sections, and formal models (github.com/block/buzz).
- **Block context — not to be missed.** This isn't an isolated move: Block already has goose and the MCP-UI work (block-goose-mcp-ui-future-agentic-interfaces-2025-08-25), and its stance on **AI monetization** is documented elsewhere (paymentsdive-block-dorsey-pricing-ia-2026-08-06). The post's self-description — *« That's what an intelligence company like Block needs »* — is worth noting: **Block calls itself an *intelligence company***, not a payments company.
- **What the post doesn't say, and should be asked**: no figures (adoption, productivity, cost), no comparison to Slack/Teams/existing forge, no discussion of the **operational cost** of self-hosting, nor of the channel moderation/governance model at enterprise scale. The formal proofs cover **storage**, not the claimed benefits.

## RésuméDe400mots

Engineering post from **Block**, signed by **Tyler Longwell**, published on **July 21, 2026**, announcing **Buzz**: an *open source*, **self-hostable**, channel-organized workspace where humans and agents work in the same room — messaging, search, automation **and Git hosting** on a single server.

**The starting point is a lived failure.** The author built Block's first Slack agent; it worked, but left operational questions unanswered: does everyone get their own bot? If a bot is shared, **whose credentials are they**? What happens when a team switches model or *runtime*? Hence the thesis: *« Models can do the work now. Teams still need somewhere to do it together. The bottleneck moved from intelligence to coordination. »*

**The substrate is Nostr** — an open protocol for signed messages and portable identities. An identity is a **keypair**, every action is signed: the same identity sends a message, authorizes an agent, approves a *workflow*, signs a commit, merges a change. **Claude Code, Codex, goose, and any agent speaking Agent Client Protocol** work inside Buzz; switching model or *harness* leaves the project's identity, permissions, and history intact.

**The core of the post is agent identity.** Rather than lending one's credentials to a bot — *« We have been letting bots play dress-up as us »* — each agent gets **its own key**. Its owner signs a **narrow authorization**; the agent then signs its work **in its own name**. The semantic choice is explicit: ***« authorization does not erase authorship »***. A compromised agent's key is revoked **without touching the human identity**; withdrawing the owner disconnects the agent.

**Second engineering piece: Git on object storage.** Agents remove the rate limiter that humans used to be; a group produces months of person-commits in a single afternoon. Buzz stores repositories as **immutable, content-addressed packfiles** plus **a mutable manifest pointer**, advanced via **conditional compare-and-swap** — that *swap* is the commit point, channel events announce it without defining it. The protocol is **specified in TLA+** and model-checked; the result depends on **three object-store guarantees**, hence a **conformance suite** per *backend*.

**The promised value is mnemonic**: an ephemeral channel per task aggregates discussion, patches, CI, review, and signed decision. *« A conventional forge preserves the diff and a green check. Buzz also preserves why the obvious fix was wrong. »*

**And open source is argued for**: *« it's 2026: software got cheap. Taste didn't. »* If Buzz disappears, identities and signed history still verify. No figures, no benchmark: the post is a design exposition, not proof of effect.

## GrapheDeConnaissance

- Block —publie→ Buzz (TECHNOLOGIE, 0.98)
- Tyler Longwell —travaille_chez→ Block (ORGANISATION, 0.97)
- Tyler Longwell —affirme_que→ le goulet d'étranglement est passé de l'intelligence à la coordination : les modèles savent faire le travail, les équipes ont encore besoin d'un lieu pour le faire ensemble (CITATION, 0.96)
- Buzz —utilise→ Nostr (TECHNOLOGIE, 0.97)
- Nostr —permet→ des identités portables sous forme de paires de clés et des actions signées, indépendantes du serveur qui les héberge (AFFIRMATION, 0.95)
- Buzz —permet→ à chaque agent d'avoir sa propre clé, son autorisation signée par son propriétaire et sa propre signature sur son travail (AFFIRMATION, 0.97)
- Buzz —affirme_que→ l'autorisation n'efface pas la paternité : l'agent reste l'auteur, son credential prouve qui l'a autorisé et sous quelles conditions (CITATION, 0.96)
- identité déléguée d'agent —résout→ l'usage des identifiants d'un humain par un bot, et permet de révoquer un agent compromis sans remplacer l'identité humaine (AFFIRMATION, 0.95)
- Buzz —utilise→ Agent Client Protocol (TECHNOLOGIE, 0.95)
- Buzz —s_applique_à→ Claude Code, Codex et goose, ainsi qu'à tout agent parlant Agent Client Protocol (AFFIRMATION, 0.95)
- Buzz —permet→ de changer de modèle ou de harness sans que le projet perde son identité, ses permissions ou son historique (AFFIRMATION, 0.93)
- Tyler Longwell —affirme_que→ les agents ont rendu le travail individuel plus rapide et le travail d'équipe plus lent, chaque passage de main exigeant un humain pour transporter le contexte (CITATION, 0.94)
- agents de codage —s_oppose_à→ le limiteur de débit humain sur lequel les forges Git étaient implicitement dimensionnées (AFFIRMATION, 0.93)
- Buzz —utilise→ des packfiles immuables adressés par contenu sur stockage objet, plus un pointeur de manifeste mutable avancé par compare-and-swap conditionnel (AFFIRMATION, 0.96)
- Buzz —affirme_que→ la mise à jour du pointeur est le point de commit, les événements du workspace annonçant le changement sans le définir (AFFIRMATION, 0.95)
- Block —utilise→ TLA+ (TECHNOLOGIE, 0.95)
- TLA+ —permet→ de model-checker la durabilité, la reconstruction et les poussées concurrentes du protocole de stockage Git de Buzz (AFFIRMATION, 0.94)
- Buzz —utilise→ une suite de conformité que chaque backend de stockage objet doit passer, le résultat borné dépendant de trois garanties explicites (AFFIRMATION, 0.93)
- Buzz —permet→ de router les requêtes d'inférence d'un agent vers la machine d'un autre membre, le trafic chiffré circulant directement entre pairs autorisés (AFFIRMATION, 0.92)
- Buzz —réduit→ la compression du contexte partagé dans des prompts privés, en gardant discussion, patchs, CI, revue et décision de merge dans un même canal (AFFIRMATION, 0.93)
- Buzz —affirme_que→ une forge classique conserve le diff et une coche verte, là où Buzz conserve aussi pourquoi le correctif évident était faux (CITATION, 0.95)
- Buzz —concurrence→ Slack (TECHNOLOGIE, 0.82)
- Block —affirme_que→ une équipe doit contrôler ses propres identités d'agents, son historique et son travail, indépendamment de tout fournisseur ou plateforme (AFFIRMATION, 0.95)
- Buzz —permet→ à une identité et à un historique signé de rester vérifiables même si Buzz disparaît, un dépôt Git restant réhébergeable (AFFIRMATION, 0.93)
- Block —affirme_que→ en 2026 le logiciel est devenu bon marché mais pas le goût, les choix de design étant la partie difficile (CITATION, 0.92)
- Buzz —est_instance_de→ workspace collaboratif humains-agents auto-hébergeable (CONCEPT, 0.95)
- Block —publie→ goose (TECHNOLOGIE, 0.9)

---
Canonical: https://www.thekb.eu/en/fiches/longwell-block-buzz-workspace-agents-nostr-2026-07-21/